PyTorch .randn()

The torch.randn() function in PyTorch generates a tensor with random numbers drawn from a standard normal distribution (mean = 0, variance = 1). This function is particularly useful for initializing weights in neural networks and for other purposes requiring randomized data, especially in machine learning applications.

torch.randn() Syntax

torch.randn(*size, *, generator=None, out=None, dtype=None, layout=torch.strided, device=None, requires_grad=False, pin_memory=False)

Parameters:

• *size: Specifies the shape of the output tensor. This can be one or more integers, defining the dimensions of the tensor.
• generator (Optional): A pseudorandom number generator for sampling.
• out (Optional): A tensor where the result will be stored. If specified, the output is stored in this tensor; otherwise, a new tensor is created.
• dtype (Optional): The desired data type of the returned tensor.
• layout (Optional): The desired layout of the returned tensor. The default is torch.strided.
• device (Optional): The device on which the tensor is allocated.
• requires_grad (Optional): If set to True, PyTorch will track operations on the tensor for automatic differentiation. Defaults to False.
• pin_memory (Optional): If set to True, the tensor will be allocated in page-locked memory, which can speed up data transfer to CUDA devices.

Return value:

Returns a tensor of the specified shape, containing random numbers drawn from a normal distribution.

Example 1: Basic Usage of torch.randn()

This example uses torch.randn() to create a 4x4 tensor with random values from a normal distribution:

import torch

tensor = torch.randn(4, 4)

print(tensor)

Copy to clipboard
Copy to clipboard

Here is the output:

tensor([[ 1.6012,  0.1432,  2.1874,  1.1822],
        [-1.4331, -1.4752,  1.5541,  1.1185],
        [ 0.2751, -0.7885, -1.0202, -0.7851],
        [ 0.8050, -0.6548,  1.7703, -1.3595]])

Copy to clipboard
Copy to clipboard

Note: This code may generate a random output each time it is run, as torch.randn() produces random values from a normal distribution.

Example 2: Specifying Device and Data Type in torch.randn()

This example uses torch.randn() to create a 4×4 tensor on the CPU with double precision:

import torch

tensor = torch.randn(4, 4, device='cpu', dtype=torch.float64)

print(tensor)

Copy to clipboard
Copy to clipboard

Here is the output:

tensor([[-0.8092, -0.6702, -0.6960,  0.2639],
        [ 0.6808,  0.5304, -0.8820, -0.7556],
        [ 0.1226, -1.0764,  0.9428,  1.4216],
        [-0.5007,  1.3984,  0.0343, -1.1393]], dtype=torch.float64)

Copy to clipboard
Copy to clipboard

Example 3: Using requires_grad in torch.randn()

This example uses torch.randn() with requires_grad set to True to create a 4×4 tensor with gradient tracking:

import torch

tensor = torch.randn(4, 4, requires_grad=True)

print(tensor)

Copy to clipboard
Copy to clipboard

Here is the output:

tensor([[ 0.8901,  0.6525, -0.7170, -1.2236],
        [-1.7839,  0.0180, -0.8420,  0.6561],
        [ 0.1359,  0.8233, -0.4252,  0.7597],
        [-1.1085,  0.9300,  0.1958, -0.9679]], requires_grad=True)

Copy to clipboard
Copy to clipboard

Frequently Asked Questions

1. What does randn do in PyTorch?

torch.randn() creates a tensor containing random numbers drawn from a normal distribution.

2. What is the difference between torch.rand() and torch.randn()?

• torch.rand() generates numbers evenly spread between 0 and 1.
• torch.randn() generates numbers based on the bell curve, centered at 0 with most values falling between -3 and 3.

3. What is the range of torch.randn()?

Theoretically, values for torch.randn() range from negative infinity to positive infinity because of the normal distribution. However, in practice, most values are between -3 and 3.